Download Determination of the Substrate Specificities of Protein Kinases by

Determination of the Substrate Specificities of
Protein Kinases by Prediction Algorithms and
Substrate Peptide Microarrays
Sommerfeld, D., Lai, S., Safaei, J., Shi, J., Zhang, H. and Pelech, S.
1. Overview
Peptide microarrays were employed to experimentally elucidate the optimal
substrate amino acid sequence information for 112 human protein kinases to
compare with predictions from the Kinase Predictor 1.0 algorithm. These
findings will be used to generate training data to improve Kinase Predictor
2.0, to identify novel kinase-substrate interactions, and to map novel
phosphorylation-dependent signalling pathways. 2. Introduction
Protein kinases play a virtually universal role in the regulation of eukaryotic
cellular processes by phosphorylating a plethora of protein (and lipid)
substrates. Over two thirds of the proteins encoded by the human genome
are subjected to phosphorylation on multiple sites, and there may be in
excess of 650,000 phosphorylation sites (P-sites) in the human proteome. To
ensure signalling fidelity amidst the vast complement of potential substrate
P-sites, protein kinases must exert specificity for their substrates and act
only on appropriate cellular targets. Substrate recognition by a particular
kinase is largely determined by molecular recognition of the amino acid
sequence surrounding a target P-site. An important aspect of substrate
peptide specificity is that the target P-site falls within a consensus amino acid
sequence motif, which exhibits structural and chemical complementarity to
the kinase active site. Consensus P-site motifs for individual kinases have
been determined through alignment of known substrate P-sites, screening of
substrate peptide libraries in solution and arrays, and systematic mutation of
substrate sequences. However, the substrate consensus sequences for most
of the 516 human protein kinases remain unknown. To fill this gap in
knowledge, there have been several attempts to predict kinase specificities
for substrate recognition in the last decade since the sequencing of the
human genome.
3. Experimental Approach
Figure 1: Overview of input and output of Kinase Predictor 1.0
INPUT (a) Aligned amino acid sequences
for P-sites from known substrates
for 229 ‘typical’ protein kinases
(from 9,125 confirmed kinasesubstrate pairs)
(b) Aligned amino acid sequences
for 488 catalytic domains from 478
‘typical’ kinases
Determine SDRs and their interaction probabilities for each amino acid position in the P-site region of
optimal substrates using deduced consensus P-sites and kinase catalytic domain information. OUTPUT
PSSM matrices of all kinases.
PSSMs are used to derive an
optimal P-site consensus
sequence.
Algorithm 2 Prediction of PSSM matrices for all kinases
Generate Profile and PSSM matrices for each kinase catalytic domain using SDR’s, predicted amino
acid interactions, and the unique amino acid sequence of each of the 488 catalytic domains. Figure 2: Synthesis and printing of optimal peptide sequences
onto glass slide microarrays
We previously developed computer algorithms to predict the specificities for
all of the human typical protein kinases (Kinase Predictor 1.0). These
algorithms utilize mutual information and charge information theory to
identify Specificity Determining Residues (SDRs) in the catalytic domains of
kinases and predict the preferred substrate sequences (13 amino acid
residues in length) for 488 typical human protein kinase catalytic domains.
They generate a frequency table or Position Specific Scoring Matrix (PSSM)
that provides the probability of occurrence for each of the 20 amino acids at
each of the 12 positions surrounding a target P-site for a specific protein
kinase (Figure 1).
Algorithm 1 Computing SDRs PEPTIDE MICROARRAY PRINTING
Peptides printed onto epoxysilanecoated glass slides. Epoxysilane coating
provides an epoxy ring that reacts with
and covalently binds amine groups on
the spotted peptides.
Image below is modified from SCHOTT
Nexterion® SlideE Product Information Sheet.
MICROARRAY SET-UP
Every peptide microarray
has 4 fields that each
contain 445 peptides
(plus control spots and
orientation markers)
printed in triplicate.
This setup allows for the
testing of 4 different
kinases per slide.
While the accuracy of our algorithms have been tested in silico by comparing
the computer-predicted consensus sequences with deduced consensus
sequences by alignment of known protein substrates, they have yet to be
validated empirically by in vitro assays. To test the accuracy of our
algorithms, we used the PSSM for each of 488 typical human protein kinase
catalytic domains to generate a list of ‘optimal’ substrate peptide
sequences. We selected 445 peptide sequences (omitting nearly identical
consensus sequences for highly related kinases) and printed them onto
glass slide microarrays (Figure 2).
Figure 3: Detection of phosphorylation with Pro-Q
Diamond phosphoprotein stain
Pro-Q Diamond® (Invitrogen) is a proprietary fluorescent stain that binds
directly to the highly negatively charged phosphate group on
phosphorylated serine, threonine and tyrosine amino acid residues, and
allows for the detection of phosphopeptides (or phosphoproteins) without
the requirement for antibodies or radioisotopes. Image below is modified
from Gao et al., 2010.*
Peptide microarrays were incubated with ATP and purified recombinant
active kinases (SignalChem Pharmaceuticals). The phosphorylation of the
peptides was detected using the fluorescent Pro-Q Diamond®
phosphoprotein/peptide stain (Figure 3). Slides were scanned and the
images were analyzed using advanced microarray image analysis software
(ImaGene® 8.0).
*Gao, L., Sun, H. and Yao, S.Q. (2010). Activity-based high-throughput determination of
PTPs substrate specificity using a phosphopeptide microarray. Peptide Science 94 (6):
810-819.
4. Results
We are aiming to test 204 purified recombinant human protein kinases in
high-throughput phosphorylation assays against 445 peptide substrates.
To date we have evaluated 112 kinases, and these different kinases
exhibit distinct substrate fingerprints/phosphorylation profiles (Figure 4).
The Pro-Q Diamond stain binds directly to the phosphate groups on
phosphorylated peptides. Therefore, the signal intensity from peptide
spots is proportional to the degree of phosphorylation of the respective
peptide substrate, and we can use the signal intensity as a measure of the
‘efficiency’ of a peptide as a substrate for a particular kinase.
Peptide substrates are ranked based on signal intensities (or degree of
phosphorylation), and the sequences of the top ranking peptides are
aligned and used to derive consensus sequences for target kinases. Table
1 demonstrates the alignment of top ranking peptide substrates for the
AMPKα1 kinase and the comparison between the consensus sequences
obtained from our peptide arrays (Peptide Consensus), alignment of
known substrates P-sites (Protein Consensus), and our Kinase Predictor
1.0 algorithm (Predicted). Comparison of the consensus sequences
shows a close match between the different methods, indicating that our
algorithms are capable of predicting specificities of kinases, and that our
peptide array can be used to test kinase specificities and validate our
algorithms.
Our preliminary results indicate that the first generation kinase specificity
algorithms are capable of predicting optimal substrate sequences with
relatively high accuracy. Table 2 illustrates the comparison of consensus
sequences derived from our peptide arrays, computer prediction, and
alignment of confirmed substrate proteins for a variety of protein kinases . Figure 4:
Phosphorylation profile
of the kinase AMPKa1
Different kinases display distinct
substrate fingerprints. Red
indicates high signal intensity
corresponding to higher
phosphorylation, while green
indicates low signal intensity.
5. Significance and Future Directions
Table 1: AMPKα1
peptide
substrate
ranking and
alignment. Peptides exhibiting
signal intensities
above 20% of the
highest signal
detected are aligned
and used to derive a
consensus sequence
(Peptide Consensus). Table 2: Comparison of
consensus sequences obtained
from peptide arrays (Peptide
Consensus), Kinase Predictor 1.0
algorithm (Predicted), and
alignment of known substrates Psites (Protein Consensus)
Shown are the comparisons of derived
consensus sequences for protein kinases of
various groups and families. Our results also
demonstrate the similarities in substrate
specificities between members of the same
kinase groups.
The substrate phosphorylation data obtained from testing all of 204 protein
kinases against our peptide microarrays will be used to expand the input
datasets for training of our algorithms. In addition, the acquisition of 3D
structural information of kinase catalytic domains, the incorporation of a
larger number of SDRs, and the adoption of an empirically-derived amino
acid interaction table will also be included in the development of second
generation Kinase Predictor algorithms with improved accuracy and
predictive power for mapping phosphoproteome interactions.
Importantly, the in silico prediction of kinase specificities and substrate
interactions can significantly facilitate the identification of kinase-substrate
interactions in vivo.
Understanding the molecular determinants that guide substrate recognition
by a kinase is essential to understanding the role of individual protein kinases
in particular cellular processes. The knowledge of substrate specificities that
can be gained through our work has significant implications to the mapping of
phosphorylation sites in the human proteome, the identification of previously
uncharacterized substrates, and the generation of model substrates for smallmolecule inhibitor design for drug development. The identification of novel Psites and substrates for kinases may ultimately lead to a high-resolution map
of kinase-dependent phosphorylation signalling networks inside the cell, and
a greater understanding of their role in the pathology of many human
diseases, including cancer, diabetes, neurodegenerative and immune
disorders.